Introduce multiparam gridding

[JaLuka98]

This pull request significantly extends the Neyman construction and profiling utilities for multi-parameter statistical inference in the npkit package. The main focus is on supporting grid-based and profiled grid-based confidence belt construction, with new helper functions and data structures for efficient and flexible multi-parameter scans.

Key additions and improvements:

Multi-parameter grid support

• Introduced the GridBelt dataclass to represent critical quantiles for multi-parameter grid scans, enabling more general Neyman belt construction beyond 1D cases.
• Added build_grid_belt for constructing Neyman belts over arbitrary parameter grids without fitting, and build_profiled_grid_belt for building 1D belts by profiling over a grid of nuisance parameters.

Vectorized and utility functions

• Implemented precompute_predictions for efficient batch evaluation of model predictions on parameter grids, and chi2_grid for vectorized computation of chi-squared statistics over grids.
• Added q_grid_profile and q_grid_profile_slice for evaluating profile likelihood-ratio statistics on grids, supporting both fixed-slice and best-fit approaches.
• Introduced grid_mask_for_fixed_params to select grid points matching fixed parameter values, facilitating profiling and slicing operations.

API and usability enhancements

• Extended build_belts_from_grid to support batch processing via a batch_size parameter and improved quantile extraction for multiple confidence levels in a single pass. [1] [2] [3]
• Updated __init__.py to expose all new functions and the GridBelt class in the public API. [1] [2]

These changes provide a robust foundation for multi-parameter confidence interval construction, profiling, and efficient statistical computation workflows in the package.